Design of Collision Warning System Based on the Fusion of In-vehicle and Out-of-vehicle Information

Recent years have witnessed a tremendous growth in the number of traffic accidents, resulting in numerous pedestrian fatalities and substantial economic losses. Collision warning systems have shown potential in mitigating such incidents by alerting drivers to imminent risks. However, existing systems often fail to consider critical human factors like driver fatigue, which impair drivers’ ability to respond effectively to warnings. To address this limitation, this study introduces a collision warning system that simultaneously evaluates road conditions and driver status. By leveraging the You Only Look Once (YOLO) algorithm for road condition analysis alongside a fatigue detection algorithm based on behavioral characteristics of the eye closure percentage (PERCLOS) and mouth closure percentage (FOM) for driver fatigue analysis, the system integrates data from onboard sensors such as cameras and assesses collision risks through fuzzy inference in real-time. Alerts are delivered through a Human-Machine Interface (HMI) to optimize driver response. This paper uses the Common Objects in Context (COCO) Dataset and the NTHU-Drowsy Driver Dataset (NTHU-DDD) to train the model. We conducted field experiments, selecting 40 sets of scenario data for model testing and recruiting 20 experienced drivers for risk annotation. The experimental results indicate that the accuracy of the proposed model achieves 86.7% in collision risk prediction, outperforming traditional systems in adaptability and safety across diverse driving scenarios. These findings emphasize that collision warning system considering driver status can provide more appropriate warning strategies and significantly enhance road safety.